[X86] Use __builtin_convertvector to implement some of the packed integer to packed float conversion intrinsics.

I believe this is safe assuming default default FP environment. The conversion might be inexact, but it can never overflow the FP type so this shouldn't be undefined behavior for the uitofp/sitofp instructions.

We already do something similar for scalar conversions.

Differential Revision: https://reviews.llvm.org/D46863

llvm-svn: 332882

6 files changed, 72 insertions, 108 deletions

diff --git a/clang/lib/Headers/avx512dqintrin.h b/clang/lib/Headers/avx512dqintrin.h
index 2c431d9740c..e75c958434a 100644
--- a/clang/lib/Headers/avx512dqintrin.h
+++ b/clang/lib/Headers/avx512dqintrin.h
@@ -361,26 +361,21 @@ _mm512_maskz_cvtps_epu64 (__mmask8 __U, __m256 __A) {
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_cvtepi64_pd (__m512i __A) {
-  return (__m512d) __builtin_ia32_cvtqq2pd512_mask ((__v8di) __A,
-                (__v8df) _mm512_setzero_pd(),
-                (__mmask8) -1,
-                _MM_FROUND_CUR_DIRECTION);
+  return (__m512d)__builtin_convertvector((__v8di)__A, __v8df);
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_mask_cvtepi64_pd (__m512d __W, __mmask8 __U, __m512i __A) {
-  return (__m512d) __builtin_ia32_cvtqq2pd512_mask ((__v8di) __A,
-                (__v8df) __W,
-                (__mmask8) __U,
-                _MM_FROUND_CUR_DIRECTION);
+  return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U,
+                                              (__v8df)_mm512_cvtepi64_pd(__A),
+                                              (__v8df)__W);
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtepi64_pd (__mmask8 __U, __m512i __A) {
-  return (__m512d) __builtin_ia32_cvtqq2pd512_mask ((__v8di) __A,
-                (__v8df) _mm512_setzero_pd(),
-                (__mmask8) __U,
-                _MM_FROUND_CUR_DIRECTION);
+  return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U,
+                                              (__v8df)_mm512_cvtepi64_pd(__A),
+                                              (__v8df)_mm512_setzero_pd());
 }
 
 #define _mm512_cvt_roundepi64_pd(A, R) __extension__ ({          \
@@ -596,26 +591,21 @@ _mm512_maskz_cvttps_epu64 (__mmask8 __U, __m256 __A) {
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_cvtepu64_pd (__m512i __A) {
-  return (__m512d) __builtin_ia32_cvtuqq2pd512_mask ((__v8di) __A,
-                 (__v8df) _mm512_setzero_pd(),
-                 (__mmask8) -1,
-                 _MM_FROUND_CUR_DIRECTION);
+  return (__m512d)__builtin_convertvector((__v8du)__A, __v8df);
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_mask_cvtepu64_pd (__m512d __W, __mmask8 __U, __m512i __A) {
-  return (__m512d) __builtin_ia32_cvtuqq2pd512_mask ((__v8di) __A,
-                 (__v8df) __W,
-                 (__mmask8) __U,
-                 _MM_FROUND_CUR_DIRECTION);
+  return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U,
+                                              (__v8df)_mm512_cvtepu64_pd(__A),
+                                              (__v8df)__W);
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtepu64_pd (__mmask8 __U, __m512i __A) {
-  return (__m512d) __builtin_ia32_cvtuqq2pd512_mask ((__v8di) __A,
-                 (__v8df) _mm512_setzero_pd(),
-                 (__mmask8) __U,
-                 _MM_FROUND_CUR_DIRECTION);
+  return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U,
+                                              (__v8df)_mm512_cvtepu64_pd(__A),
+                                              (__v8df)_mm512_setzero_pd());
 }
 
 #define _mm512_cvt_roundepu64_pd(A, R) __extension__ ({          \
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 92d1ad26a94..b4cfe09a426 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -3831,28 +3831,23 @@ _mm512_maskz_cvttps_epu32 (__mmask16 __U, __m512 __A)
 static __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_cvtepu32_ps (__m512i __A)
 {
-  return (__m512) __builtin_ia32_cvtudq2ps512_mask ((__v16si) __A,
-                 (__v16sf) _mm512_undefined_ps (),
-                 (__mmask16) -1,
-                 _MM_FROUND_CUR_DIRECTION);
+  return (__m512)__builtin_convertvector((__v16su)__A, __v16sf);
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_mask_cvtepu32_ps (__m512 __W, __mmask16 __U, __m512i __A)
 {
-  return (__m512) __builtin_ia32_cvtudq2ps512_mask ((__v16si) __A,
-                 (__v16sf) __W,
-                 (__mmask16) __U,
-                 _MM_FROUND_CUR_DIRECTION);
+  return (__m512)__builtin_ia32_selectps_512((__mmask16)__U,
+                                             (__v16sf)_mm512_cvtepu32_ps(__A),
+                                             (__v16sf)__W);
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtepu32_ps (__mmask16 __U, __m512i __A)
 {
-  return (__m512) __builtin_ia32_cvtudq2ps512_mask ((__v16si) __A,
-                 (__v16sf) _mm512_setzero_ps (),
-                 (__mmask16) __U,
-                 _MM_FROUND_CUR_DIRECTION);
+  return (__m512)__builtin_ia32_selectps_512((__mmask16)__U,
+                                             (__v16sf)_mm512_cvtepu32_ps(__A),
+                                             (__v16sf)_mm512_setzero_ps());
 }
 
 static __inline __m512d __DEFAULT_FN_ATTRS
@@ -3892,28 +3887,23 @@ _mm512_mask_cvtepi32lo_pd(__m512d __W, __mmask8 __U,__m512i __A)
 static __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_cvtepi32_ps (__m512i __A)
 {
-  return (__m512) __builtin_ia32_cvtdq2ps512_mask ((__v16si) __A,
-                (__v16sf) _mm512_undefined_ps (),
-                (__mmask16) -1,
-                _MM_FROUND_CUR_DIRECTION);
+  return (__m512)__builtin_convertvector((__v16si)__A, __v16sf);
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_mask_cvtepi32_ps (__m512 __W, __mmask16 __U, __m512i __A)
 {
-  return (__m512) __builtin_ia32_cvtdq2ps512_mask ((__v16si) __A,
-                (__v16sf) __W,
-                (__mmask16) __U,
-                _MM_FROUND_CUR_DIRECTION);
+  return (__m512)__builtin_ia32_selectps_512((__mmask16)__U,
+                                             (__v16sf)_mm512_cvtepi32_ps(__A),
+                                             (__v16sf)__W);
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtepi32_ps (__mmask16 __U, __m512i __A)
 {
-  return (__m512) __builtin_ia32_cvtdq2ps512_mask ((__v16si) __A,
-                (__v16sf) _mm512_setzero_ps (),
-                (__mmask16) __U,
-                _MM_FROUND_CUR_DIRECTION);
+  return (__m512)__builtin_ia32_selectps_512((__mmask16)__U,
+                                             (__v16sf)_mm512_cvtepi32_ps(__A),
+                                             (__v16sf)_mm512_setzero_ps());
 }
 
 static __inline __m512d __DEFAULT_FN_ATTRS
diff --git a/clang/lib/Headers/avx512vldqintrin.h b/clang/lib/Headers/avx512vldqintrin.h
index d80df9eaffe..e1b9abb25c0 100644
--- a/clang/lib/Headers/avx512vldqintrin.h
+++ b/clang/lib/Headers/avx512vldqintrin.h
@@ -463,44 +463,40 @@ _mm256_maskz_cvtps_epu64 (__mmask8 __U, __m128 __A) {
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_cvtepi64_pd (__m128i __A) {
-  return (__m128d) __builtin_ia32_cvtqq2pd128_mask ((__v2di) __A,
-                (__v2df) _mm_setzero_pd(),
-                (__mmask8) -1);
+  return (__m128d)__builtin_convertvector((__v2di)__A, __v2df);
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_mask_cvtepi64_pd (__m128d __W, __mmask8 __U, __m128i __A) {
-  return (__m128d) __builtin_ia32_cvtqq2pd128_mask ((__v2di) __A,
-                (__v2df) __W,
-                (__mmask8) __U);
+  return (__m128d)__builtin_ia32_selectpd_128((__mmask8)__U,
+                                              (__v2df)_mm_cvtepi64_pd(__A),
+                                              (__v2df)__W);
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_maskz_cvtepi64_pd (__mmask8 __U, __m128i __A) {
-  return (__m128d) __builtin_ia32_cvtqq2pd128_mask ((__v2di) __A,
-                (__v2df) _mm_setzero_pd(),
-                (__mmask8) __U);
+  return (__m128d)__builtin_ia32_selectpd_128((__mmask8)__U,
+                                              (__v2df)_mm_cvtepi64_pd(__A),
+                                              (__v2df)_mm_setzero_pd());
 }
 
 static __inline__ __m256d __DEFAULT_FN_ATTRS
 _mm256_cvtepi64_pd (__m256i __A) {
-  return (__m256d) __builtin_ia32_cvtqq2pd256_mask ((__v4di) __A,
-                (__v4df) _mm256_setzero_pd(),
-                (__mmask8) -1);
+  return (__m256d)__builtin_convertvector((__v4di)__A, __v4df);
 }
 
 static __inline__ __m256d __DEFAULT_FN_ATTRS
 _mm256_mask_cvtepi64_pd (__m256d __W, __mmask8 __U, __m256i __A) {
-  return (__m256d) __builtin_ia32_cvtqq2pd256_mask ((__v4di) __A,
-                (__v4df) __W,
-                (__mmask8) __U);
+  return (__m256d)__builtin_ia32_selectpd_256((__mmask8)__U,
+                                              (__v4df)_mm256_cvtepi64_pd(__A),
+                                              (__v4df)__W);
 }
 
 static __inline__ __m256d __DEFAULT_FN_ATTRS
 _mm256_maskz_cvtepi64_pd (__mmask8 __U, __m256i __A) {
-  return (__m256d) __builtin_ia32_cvtqq2pd256_mask ((__v4di) __A,
-                (__v4df) _mm256_setzero_pd(),
-                (__mmask8) __U);
+  return (__m256d)__builtin_ia32_selectpd_256((__mmask8)__U,
+                                              (__v4df)_mm256_cvtepi64_pd(__A),
+                                              (__v4df)_mm256_setzero_pd());
 }
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS
@@ -715,44 +711,40 @@ _mm256_maskz_cvttps_epu64 (__mmask8 __U, __m128 __A) {
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_cvtepu64_pd (__m128i __A) {
-  return (__m128d) __builtin_ia32_cvtuqq2pd128_mask ((__v2di) __A,
-                (__v2df) _mm_setzero_pd(),
-                (__mmask8) -1);
+  return (__m128d)__builtin_convertvector((__v2du)__A, __v2df);
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_mask_cvtepu64_pd (__m128d __W, __mmask8 __U, __m128i __A) {
-  return (__m128d) __builtin_ia32_cvtuqq2pd128_mask ((__v2di) __A,
-                (__v2df) __W,
-                (__mmask8) __U);
+  return (__m128d)__builtin_ia32_selectpd_128((__mmask8)__U,
+                                              (__v2df)_mm_cvtepu64_pd(__A),
+                                              (__v2df)__W);
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_maskz_cvtepu64_pd (__mmask8 __U, __m128i __A) {
-  return (__m128d) __builtin_ia32_cvtuqq2pd128_mask ((__v2di) __A,
-                (__v2df) _mm_setzero_pd(),
-                (__mmask8) __U);
+  return (__m128d)__builtin_ia32_selectpd_128((__mmask8)__U,
+                                              (__v2df)_mm_cvtepu64_pd(__A),
+                                              (__v2df)_mm_setzero_pd());
 }
 
 static __inline__ __m256d __DEFAULT_FN_ATTRS
 _mm256_cvtepu64_pd (__m256i __A) {
-  return (__m256d) __builtin_ia32_cvtuqq2pd256_mask ((__v4di) __A,
-                (__v4df) _mm256_setzero_pd(),
-                (__mmask8) -1);
+  return (__m256d)__builtin_convertvector((__v4du)__A, __v4df);
 }
 
 static __inline__ __m256d __DEFAULT_FN_ATTRS
 _mm256_mask_cvtepu64_pd (__m256d __W, __mmask8 __U, __m256i __A) {
-  return (__m256d) __builtin_ia32_cvtuqq2pd256_mask ((__v4di) __A,
-                (__v4df) __W,
-                (__mmask8) __U);
+  return (__m256d)__builtin_ia32_selectpd_256((__mmask8)__U,
+                                              (__v4df)_mm256_cvtepu64_pd(__A),
+                                              (__v4df)__W);
 }
 
 static __inline__ __m256d __DEFAULT_FN_ATTRS
 _mm256_maskz_cvtepu64_pd (__mmask8 __U, __m256i __A) {
-  return (__m256d) __builtin_ia32_cvtuqq2pd256_mask ((__v4di) __A,
-                (__v4df) _mm256_setzero_pd(),
-                (__mmask8) __U);
+  return (__m256d)__builtin_ia32_selectpd_256((__mmask8)__U,
+                                              (__v4df)_mm256_cvtepu64_pd(__A),
+                                              (__v4df)_mm256_setzero_pd());
 }
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS
diff --git a/clang/lib/Headers/avx512vlintrin.h b/clang/lib/Headers/avx512vlintrin.h
index 213d911fd5d..9e03b20a3ad 100644
--- a/clang/lib/Headers/avx512vlintrin.h
+++ b/clang/lib/Headers/avx512vlintrin.h
@@ -2207,48 +2207,40 @@ _mm256_maskz_cvtepu32_pd (__mmask8 __U, __m128i __A) {
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS
 _mm_cvtepu32_ps (__m128i __A) {
-  return (__m128) __builtin_ia32_cvtudq2ps128_mask ((__v4si) __A,
-                (__v4sf)
-                _mm_setzero_ps (),
-                (__mmask8) -1);
+  return (__m128)__builtin_convertvector((__v4su)__A, __v4sf);
 }
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS
 _mm_mask_cvtepu32_ps (__m128 __W, __mmask8 __U, __m128i __A) {
-  return (__m128) __builtin_ia32_cvtudq2ps128_mask ((__v4si) __A,
-                (__v4sf) __W,
-                (__mmask8) __U);
+  return (__m128)__builtin_ia32_selectps_128((__mmask8)__U,
+                                             (__v4sf)_mm_cvtepu32_ps(__A),
+                                             (__v4sf)__W);
 }
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS
 _mm_maskz_cvtepu32_ps (__mmask8 __U, __m128i __A) {
-  return (__m128) __builtin_ia32_cvtudq2ps128_mask ((__v4si) __A,
-                (__v4sf)
-                _mm_setzero_ps (),
-                (__mmask8) __U);
+  return (__m128)__builtin_ia32_selectps_128((__mmask8)__U,
+                                             (__v4sf)_mm_cvtepu32_ps(__A),
+                                             (__v4sf)_mm_setzero_ps());
 }
 
 static __inline__ __m256 __DEFAULT_FN_ATTRS
 _mm256_cvtepu32_ps (__m256i __A) {
-  return (__m256) __builtin_ia32_cvtudq2ps256_mask ((__v8si) __A,
-                (__v8sf)
-                _mm256_setzero_ps (),
-                (__mmask8) -1);
+  return (__m256)__builtin_convertvector((__v8su)__A, __v8sf);
 }
 
 static __inline__ __m256 __DEFAULT_FN_ATTRS
 _mm256_mask_cvtepu32_ps (__m256 __W, __mmask8 __U, __m256i __A) {
-  return (__m256) __builtin_ia32_cvtudq2ps256_mask ((__v8si) __A,
-                (__v8sf) __W,
-                (__mmask8) __U);
+  return (__m256)__builtin_ia32_selectps_256((__mmask8)__U,
+                                             (__v8sf)_mm256_cvtepu32_ps(__A),
+                                             (__v8sf)__W);
 }
 
 static __inline__ __m256 __DEFAULT_FN_ATTRS
 _mm256_maskz_cvtepu32_ps (__mmask8 __U, __m256i __A) {
-  return (__m256) __builtin_ia32_cvtudq2ps256_mask ((__v8si) __A,
-                (__v8sf)
-                _mm256_setzero_ps (),
-                (__mmask8) __U);
+  return (__m256)__builtin_ia32_selectps_256((__mmask8)__U,
+                                             (__v8sf)_mm256_cvtepu32_ps(__A),
+                                             (__v8sf)_mm256_setzero_ps());
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
diff --git a/clang/lib/Headers/avxintrin.h b/clang/lib/Headers/avxintrin.h
index b77f2e99008..ace670fcda4 100644
--- a/clang/lib/Headers/avxintrin.h
+++ b/clang/lib/Headers/avxintrin.h
@@ -2225,7 +2225,7 @@ _mm256_cvtepi32_pd(__m128i __a)
 static __inline __m256 __DEFAULT_FN_ATTRS
 _mm256_cvtepi32_ps(__m256i __a)
 {
-  return (__m256)__builtin_ia32_cvtdq2ps256((__v8si) __a);
+  return (__m256)__builtin_convertvector((__v8si)__a, __v8sf);
 }
 
 /// Converts a 256-bit vector of [4 x double] into a 128-bit vector of
diff --git a/clang/lib/Headers/emmintrin.h b/clang/lib/Headers/emmintrin.h
index 8972e8c7789..fccb91bbb71 100644
--- a/clang/lib/Headers/emmintrin.h
+++ b/clang/lib/Headers/emmintrin.h
@@ -3424,7 +3424,7 @@ _mm_cvttsd_si64(__m128d __a)
 static __inline__ __m128 __DEFAULT_FN_ATTRS
 _mm_cvtepi32_ps(__m128i __a)
 {
-  return __builtin_ia32_cvtdq2ps((__v4si)__a);
+  return (__m128)__builtin_convertvector((__v4si)__a, __v4sf);
 }
 
 /// Converts a vector of [4 x float] into a vector of [4 x i32].